Debian Codegraph @

tests/pac/trend-component-18/example1.mod

raw
1 // --+ options: json=compute, stochastic +--
2
3 var x1 x2 x1bar x2bar z y x;
4
5 varexo ex1 ex2 ex1bar ex2bar ez ey ex;
6
7 parameters
8 rho_1 rho_2 rho_3 rho_4
9 a_x1_0 a_x1_1 a_x1_2 a_x1_x2_1 a_x1_x2_2
10 a_x2_0 a_x2_1 a_x2_2 a_x2_x1_1 a_x2_x1_2
11 e_c_m c_z_1 c_z_2 beta
12 lambda;
13
14 rho_1 = .9;
15 rho_2 = -.2;
16 rho_3 = .4;
17 rho_4 = -.3;
18
19
20 a_x1_0 = -.9;
21 a_x1_1 = .4;
22 a_x1_2 = .3;
23 a_x1_x2_1 = .1;
24 a_x1_x2_2 = .2;
25
26 a_x2_0 = -.9;
27 a_x2_1 = .2;
28 a_x2_2 = -.1;
29 a_x2_x1_1 = -.1;
30 a_x2_x1_2 = .2;
31
32 beta = .2;
33 e_c_m = .5;
34 c_z_1 = .2;
35 c_z_2 = -.1;
36
37 lambda = 0.5; // Share of optimizing agents.
38
39 trend_component_model(model_name=toto, eqtags=['eq:x1', 'eq:x2', 'eq:x1bar', 'eq:x2bar'], targets=['eq:x1bar', 'eq:x2bar']);
40
41 pac_model(auxiliary_model_name=toto, discount=beta, model_name=pacman);
42
43 model;
44
45 [name='eq:y']
46 y = rho_1*y(-1) + rho_2*y(-2) + ey;
47
48 [name='eq:x']
49 x = rho_3*x(-1) + rho_4*x(-2) + ex;
50
51 [name='eq:x1']
52 diff(x1) = a_x1_0*(x1(-1)-x1bar(-1)) + a_x1_1*diff(x1(-1)) + a_x1_2*diff(x1(-2)) + a_x1_x2_1*diff(x2(-1)) + a_x1_x2_2*diff(x2(-2)) + ex1;
53
54 [name='eq:x2']
55 diff(x2) = a_x2_0*(x2(-1)-x2bar(-1)) + a_x2_1*diff(x1(-1)) + a_x2_2*diff(x1(-2)) + a_x2_x1_1*diff(x2(-1)) + a_x2_x1_2*diff(x2(-2)) + ex2;
56
57 [name='eq:x1bar']
58 x1bar = x1bar(-1) + ex1bar;
59
60 [name='eq:x2bar']
61 x2bar = x2bar(-1) + ex2bar;
62
63 [name='zpac']
64 diff(z) = lambda*(e_c_m*(x1(-1)-z(-1)) + c_z_1*diff(z(-1)) + c_z_2*diff(z(-2)) + pac_expectation(pacman)) + (1-lambda)*( y + x) + ez;
65
66 end;
67
68 shocks;
69 var ex1 = 1.0;
70 var ex2 = 1.0;
71 var ex1bar = 1.0;
72 var ex2bar = 1.0;
73 var ez = 1.0;
74 var ey = 0.1;
75 var ex = 0.1;
76 end;
77
78 // Initialize the PAC model (build the Companion VAR representation for the auxiliary model).
79 pac.initialize('pacman');
80
81 // Update the parameters of the PAC expectation model (h0 and h1 vectors).
82 pac.update.expectation('pacman');
83
84 // Set initial conditions to zero. Please use more sensible values if any...
85 initialconditions = dseries(zeros(10, M_.endo_nbr+M_.exo_nbr), 2000Q1, vertcat(M_.endo_names,M_.exo_names));
86
87 // Simulate the model for 500 periods
88 TrueData = simul_backward_model(initialconditions, 300);
89
90 // Define a structure describing the parameters to be estimated (with initial conditions).
91 clear eparams
92 eparams.e_c_m = .9;
93 eparams.c_z_1 = .5;
94 eparams.c_z_2 = .2;
95 eparams.lambda = .7;
96
97 // Define the dataset used for estimation
98 edata = TrueData;
99 edata.ez = dseries(NaN(TrueData.nobs, 1), 2000Q1, 'ez');
100
101 tic
102 pac.estimate.nls('zpac', eparams, edata, 2005Q1:2005Q1+200, 'csminwel', 'verbosity', 0);
103 toc
104 skipline(1)
105
106 e_c_m_nls = M_.params(strmatch('e_c_m', M_.param_names, 'exact'));
107 c_z_1_nls = M_.params(strmatch('c_z_1', M_.param_names, 'exact'));
108 c_z_2_nls = M_.params(strmatch('c_z_2', M_.param_names, 'exact'));
109 lambda_nls = M_.params(strmatch('lambda', M_.param_names, 'exact'));
110
111 disp(sprintf('Estimate of e_c_m: %f', e_c_m_nls))
112 disp(sprintf('Estimate of c_z_1: %f', c_z_1_nls))
113 disp(sprintf('Estimate of c_z_2: %f', c_z_2_nls))
114 disp(sprintf('Estimate of lambda: %f', lambda_nls))
115
116 skipline(2)
117
118 // Define a structure describing the parameters to be estimated (with initial conditions).
119 clear eparams
120 eparams.e_c_m = .9;
121 eparams.c_z_1 = .5;
122 eparams.c_z_2 = .2;
123 eparams.lambda = .0;
124
125 // Define the dataset used for estimation
126 edata = TrueData;
127 edata.ez = dseries(NaN(TrueData.nobs, 1), 2000Q1, 'ez');
128
129 tic
130 pac.estimate.nls('zpac', eparams, edata, 2005Q1:2005Q1+200, 'lsqnonlin', 'Algorithm', 'levenberg-marquardt');
131 toc
132
133 skipline(1)
134
135 e_c_m_lsqnonlin = M_.params(strmatch('e_c_m', M_.param_names, 'exact'));
136 c_z_1_lsqnonlin = M_.params(strmatch('c_z_1', M_.param_names, 'exact'));
137 c_z_2_lsqnonlin= M_.params(strmatch('c_z_2', M_.param_names, 'exact'));
138 lambda_lsqnonlin = M_.params(strmatch('lambda', M_.param_names, 'exact'));
139
140 disp(sprintf('Estimate of e_c_m: %f', e_c_m_lsqnonlin))
141 disp(sprintf('Estimate of c_z_1: %f', c_z_1_lsqnonlin))
142 disp(sprintf('Estimate of c_z_2: %f', c_z_2_lsqnonlin))
143 disp(sprintf('Estimate of lambda: %f', lambda_lsqnonlin))
144
145 if abs(e_c_m_nls-e_c_m_lsqnonlin)>.01
146 error('Gauss Newton and direct SSR minimization do not provide consistent estimates (e_c_m)')
147 end
148
149 if abs(c_z_1_nls-c_z_1_lsqnonlin)>.01
150 error('Gauss Newton and direct SSR minimization do not provide consistent estimates (c_z_1)')
151 end
152
153 if abs(c_z_2_nls-c_z_2_lsqnonlin)>.01
154 error('Gauss Newton and direct SSR minimization do not provide consistent estimates (c_z_2)')
155 end
156
157 if abs(lambda_nls-lambda_lsqnonlin)>.01
158 error('Gauss Newton and direct SSR minimization do not provide consistent estimates (lambda)')
159 end